{"id":7508,"date":"2026-09-01T15:00:09","date_gmt":"2026-09-01T07:00:09","guid":{"rendered":"https:\/\/1ow.top\/index.php\/2026\/09\/01\/%e8%87%aa%e8%bf%9b%e5%8c%96wam%e6%9d%a5%e4%ba%86%ef%bc%81%e6%b8%85%e5%8d%8eair%e8%81%94%e6%89%8b%e5%9f%9f%e5%8f%98%e6%8d%a2%e6%8f%90%e5%87%ba%e5%85%b7%e8%ba%abin-context-causal-learning\/"},"modified":"2026-09-01T15:00:09","modified_gmt":"2026-09-01T07:00:09","slug":"%e8%87%aa%e8%bf%9b%e5%8c%96wam%e6%9d%a5%e4%ba%86%ef%bc%81%e6%b8%85%e5%8d%8eair%e8%81%94%e6%89%8b%e5%9f%9f%e5%8f%98%e6%8d%a2%e6%8f%90%e5%87%ba%e5%85%b7%e8%ba%abin-context-causal-learning","status":"publish","type":"post","link":"https:\/\/1ow.top\/index.php\/2026\/09\/01\/%e8%87%aa%e8%bf%9b%e5%8c%96wam%e6%9d%a5%e4%ba%86%ef%bc%81%e6%b8%85%e5%8d%8eair%e8%81%94%e6%89%8b%e5%9f%9f%e5%8f%98%e6%8d%a2%e6%8f%90%e5%87%ba%e5%85%b7%e8%ba%abin-context-causal-learning\/","title":{"rendered":"\u81ea\u8fdb\u5316WAM\u6765\u4e86\uff01\u6e05\u534eAIR\u8054\u624b\u57df\u53d8\u6362\u63d0\u51fa\u5177\u8eabIn-Context Causal Learning"},"content":{"rendered":"
\"\u81ea\u8fdb\u5316WAM\u6765\u4e86\uff01\u6e05\u534eAIR\u8054\u624b\u57df\u53d8\u6362\u63d0\u51fa\u5177\u8eabIn-Context<\/div><\/figure>\n
\"\u81ea\u8fdb\u5316WAM\u6765\u4e86\uff01\u6e05\u534eAIR\u8054\u624b\u57df\u53d8\u6362\u63d0\u51fa\u5177\u8eabIn-Context<\/div><\/figure>\n
\"\u81ea\u8fdb\u5316WAM\u6765\u4e86\uff01\u6e05\u534eAIR\u8054\u624b\u57df\u53d8\u6362\u63d0\u51fa\u5177\u8eabIn-Context<\/div><\/figure>\n
\"\u81ea\u8fdb\u5316WAM\u6765\u4e86\uff01\u6e05\u534eAIR\u8054\u624b\u57df\u53d8\u6362\u63d0\u51fa\u5177\u8eabIn-Context<\/div><\/figure>\n

Zeva\u56e2\u961f \u6295\u7a3f<\/p>\n

\u6a21\u578b\u4e0d\u66f4\u65b0\uff0c\u80fd\u529b\u5374\u6301\u7eed\u589e\u957f\u3002<\/p>\n

\u8fd9\u662fZeva\u6700\u53cd\u76f4\u89c9\u7684\u5730\u65b9\uff0c\u4e5f\u662f\u6e05\u534eAIR\u4e0e\u57df\u53d8\u6362\u8054\u624b\u7ed9\u51fa\u7684\u65b0\u7b54\u6848\u3002<\/p>\n

\u5728\u591a\u4e2a\u5178\u578b\u5177\u8eab\u57fa\u51c6\u4efb\u52a1\u4e0a\uff0c\u5c06\u51bb\u7ed3\u6a21\u578b\u7684\u7d2f\u8ba1\u6210\u529f\u7387\u4ece26%\u63d0\u5347\u523073%\uff1b\u5728\u771f\u5b9e\u5316\u5b66\u5b9e\u9a8c\u5ba4\uff0c\u673a\u68b0\u81c2\u5728\u591a\u6b21\u5c1d\u8bd5\u4e2d\u8d8a\u7528\u8d8a\u7a33\u3002<\/p>\n

\u66f4\u5173\u952e\u7684\u662f\uff0c\u4e00\u6b21\u4eba\u7c7b\u6f14\u793a\u5c31\u80fd\u8ba9\u6a21\u578b\u201c\u5b66\u4f1a\u201d\u4e00\u4e2a\u65b0\u64cd\u4f5c\u3002<\/p>\n

Zeva\u662f\u9996\u4e2a\u5b9e\u73b0In-Context Causal Learning\uff08ICCL\uff09\u7684\u5177\u8eabWAM\u3002<\/p>\n

\u5b83\u8ba9\u6a21\u578b\u53ef\u4ee5\u5728\u4e0d\u66f4\u65b0\u6743\u91cd\u7684\u60c5\u51b5\u4e0b\uff0c\u4ece\u81ea\u5df1\u7684\u4ea4\u4e92\u7ecf\u9a8c\u4e2d\u6301\u7eed\u5b66\u4e60\u3002\u5f53\u8fd9\u79cd\u80fd\u529b\u9010\u6b65\u6210\u719f\uff0c\u5177\u8eab\u667a\u80fd\u5c06\u8fce\u6765\u5c5e\u4e8e\u81ea\u5df1\u7684\u201cGPT\u65f6\u523b\u201d\u3002<\/p>\n

\u9879\u76ee\u4fe1\u606f<\/p>\n

\u9879\u76ee\u540d\u79f0\uff1aZeva\u53d1\u5e03\u673a\u6784\uff1a\u6e05\u534eAIR\u3001\u57df\u53d8\u6362\u8bba\u6587\u9898\u76ee\uff1aZeva: In-Context Causal Learning for Generalizable Embodied Manipulation\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/air-embodied-brain.github.io\/Zeva\/<\/p>\n

\u5177\u8eab\u667a\u80fd\u7684\u771f\u6b63\u74f6\u9888\uff0c\u662f\u201c\u6a21\u578b\u5230\u4e86\u73b0\u573a\u80fd\u4e0d\u80fd\u8d8a\u505a\u8d8a\u597d\u201d\u3002<\/p>\n

\u4f20\u7edf\u601d\u8def\u91cc\uff0c\u60f3\u8ba9\u673a\u5668\u4eba\u9002\u5e94\u65b0\u73af\u5883\uff0c\u901a\u5e38\u8981\u91cd\u65b0\u91c7\u96c6\u6570\u636e\u3001\u5fae\u8c03\u6a21\u578b\u6216\u505atest-time training\u3002<\/p>\n

\u4f46\u8fd9\u4e9b\u65b9\u6cd5\u4f9d\u8d56\u68af\u5ea6\u66f4\u65b0\uff0c\u901f\u5ea6\u6162\u3001\u6210\u672c\u9ad8\uff0c\u8fd8\u53ef\u80fd\u5728\u9002\u5e94\u65b0\u4efb\u52a1\u65f6\u7834\u574f\u5df2\u6709\u80fd\u529b\u3002<\/p>\n

\u53e6\u4e00\u7c7bin-context learning\u65b9\u6cd5\u53ef\u4ee5\u8ba9\u6a21\u578b\u6839\u636e\u6f14\u793a\u6216\u4efb\u52a1\u63cf\u8ff0\u8c03\u6574\u884c\u4e3a\uff0c\u4f46\u4e3b\u8981\u89e3\u51b3\u201c\u4efb\u52a1\u7406\u89e3\u201d\uff0c\u5e76\u6ca1\u6709\u8ba9\u6a21\u578b\u4ece\u81ea\u5df1\u7684\u52a8\u4f5c\u540e\u679c\u4e2d\u5b66\u4e60\u7269\u7406\u56e0\u679c\u3002<\/p>\n

\u8fc7\u53bb\u6211\u4eec\u8c08\u8de8\u57df\uff0c\u66f4\u591a\u662fsim-to-real\u3001\u8de8\u672c\u4f53\u8fc1\u79fb\u3001\u8de8\u4efb\u52a1\u6cdb\u5316\u3002\u8fd9\u4e9b\u65b9\u6cd5\u5927\u591a\u4f9d\u8d56\u6570\u636e\u5bf9\u9f50\u6216\u8868\u5f81\u5bf9\u9f50\uff0c\u672c\u8d28\u4e0a\u662f\u5728\u4e0d\u540c\u57df\u4e4b\u95f4\u642c\u8fd0\u201c\u4efb\u52a1\u7ecf\u9a8c\u201d\u3002<\/p>\n

Zeva\u66f4\u8fdb\u4e00\u6b65\uff1a\u5b83\u8fc1\u79fb\u7684\u662f\u201c\u52a8\u4f5c\u4e0e\u72b6\u6001\u53d8\u5316\u4e4b\u95f4\u7684\u56e0\u679c\u5173\u7cfb\u201d\u3002<\/p>\n

\u540c\u4e00\u4e2a\u6a21\u578b\uff0c\u5728\u4eff\u771f\u91cc\u5b66\u5230\u7684\u7269\u7406\u56e0\u679c\uff0c\u53ef\u4ee5\u88ab\u5e26\u5230\u771f\u5b9e\u673a\u68b0\u81c2\uff1b\u5728\u4e00\u6b21\u5c1d\u8bd5\u91cc\u79ef\u7d2f\u7684\u5931\u8d25\u7ecf\u9a8c\uff0c\u53ef\u4ee5\u6307\u5bfc\u53e6\u4e00\u4e2a\u76f8\u4f3c\u7269\u7406\u8fc7\u7a0b\u7684\u5c1d\u8bd5\uff1b\u4eba\u7c7b\u7684\u4e00\u6b21\u6f14\u793a\uff0c\u4e5f\u53ef\u4ee5\u76f4\u63a5\u53d8\u6210\u673a\u5668\u4eba\u7684\u56e0\u679c\u4e0a\u4e0b\u6587\u3002<\/p>\n

\u8fd9\u79cd\u8de8\u57df\u8fc1\u79fb\u901a\u8fc7\u4e0a\u4e0b\u6587\u5b8c\u6210\u3002\u6a21\u578b\u5728\u4e0a\u4e0b\u6587\u4e2d\u63a8\u65ad\u73af\u5883\u56e0\u679c\u7ed3\u6784\uff0c\u5e76\u628a\u56e0\u679c\u77e5\u8bc6\u7528\u4e8e\u4e0b\u4e00\u6b21\u52a8\u4f5c\u751f\u6210\u3002\u8fd9\u5c31\u662fZeva\u6240\u5b9a\u4e49\u7684In-Context Causal Learning\u3002<\/p>\n

\u793a\u4f8b\u4e00\uff1a\u4ece\u653e\u7f6e\u79f0\u91cf\u7eb8\u7684\u8bef\u64cd\u4f5c\u4e2d\u63d0\u53d6\u56e0\u679c\u5173\u7cfb\uff0c\u5b8c\u6210\u4e0a\u4e0b\u6587\u5185\u5b66\u4e60\uff0c\u63d0\u5347\u4efb\u52a1\u7cbe\u5ea6\u3002<\/p>\n

\u793a\u4f8b\u4e8c\uff1a\u4ece\u79f0\u53d6\u8bd5\u5242\u81f3\u79f0\u91cf\u7eb8\u7684\u8bef\u64cd\u4f5c\u4e2d\u63d0\u53d6\u56e0\u679c\u5173\u7cfb\uff0c\u5b8c\u6210\u4e0a\u4e0b\u6587\u5185\u5b66\u4e60\uff0c\u63d0\u5347\u4efb\u52a1\u7cbe\u5ea6\u3002<\/p>\n

\u793a\u4f8b\u4e09\uff1a\u4ece\u4e00\u6b21\u4eba\u7c7b\u6f14\u793a\u4e2d\u5b8c\u6210\u4e0a\u4e0b\u6587\u5185\u56e0\u679c\u5b66\u4e60\u3002<\/p>\n

\u25b3Zeva\u603b\u4f53\u6846\u67b6\u3002\u6a21\u578b\u901a\u8fc7Causal Interaction Extraction\u3001Dual-timescale Causal Memory\u3001In-Context Policy Injection\uff0c\u5728\u4e0d\u66f4\u65b0\u6743\u91cd\u7684\u60c5\u51b5\u4e0b\u5b9e\u73b0\u90e8\u7f72\u4e2d\u81ea\u8fdb\u5316\u3002\u56fe\u7247\u6765\u6e90\uff1a\u8bba\u6587<\/p>\n

Zeva\u662f\u4e00\u4e2a\u6a21\u578b\u3002\u5b83\u7684Causal Transition Encoder\u3001Dual-timescale Causal Memory\u3001In-Context Policy Injection\uff0c\u90fd\u5c5e\u4e8e\u6a21\u578b\u63a8\u7406\u94fe\u8def\u7684\u4e00\u90e8\u5206\u3002<\/p>\n

\u5728\u6bcf\u4e2a\u73af\u5883\u6b65\uff0cZeva\u63d0\u53d6\u4e09\u7c7b\u4fe1\u53f7\uff1a\u89c6\u89c9\u72b6\u6001\u3001\u52a8\u4f5c\u7f16\u7801\u3001\u89c2\u5bdf\u5230\u7684\u72b6\u6001\u53d8\u5316\u3002<\/p>\n

\u5b83\u4eec\u88ab\u9012\u5f52\u7f16\u7801\u4e3aCausal Interaction State\uff0c\u5e76\u8fdb\u4e00\u6b65\u6295\u5f71\u4e3aPhase Token\u548cCausal Interaction Signal\u3002<\/p>\n

\u8fd9\u4e9b\u56e0\u679c\u4fe1\u53f7\u88ab\u7ec4\u7ec7\u6210\u53cc\u65f6\u6807\u4e0a\u4e0b\u6587\uff1aBrief Interaction Trace\u8d1f\u8d23\u5f53\u524d\u5c1d\u8bd5\u5185\u7684\u8fde\u8d2f\u6267\u884c\uff0cPersistent Interaction Memory\u8de8\u5c1d\u8bd5\u7d2f\u79ef\u53ef\u590d\u7528\u7ecf\u9a8c\u3002<\/p>\n

\u51b3\u7b56\u65f6\uff0cZeva\u68c0\u7d22\u4e0e\u5f53\u524d\u4efb\u52a1\u9636\u6bb5\u5339\u914d\u7684\u4ea4\u4e92\u8bc1\u636e\uff0c\u6784\u9020Causal Prompt\uff0c\u6ce8\u5165\u51bb\u7ed3\u7684Cosmos3\u52a8\u4f5c\u751f\u6210\u6a21\u578b\u3002\u6574\u4e2a\u8fc7\u7a0b\u96f6\u68af\u5ea6\u66f4\u65b0\u3002<\/p>\n

\u4eae\u70b9\u4e00\uff1aCausal Transition Encoder\uff0c\u8ba9\u6a21\u578b\u770b\u89c1\u201c\u52a8\u4f5c\u2192\u7ed3\u679c\u201d\u3002<\/p>\n

\u6a21\u578b\u663e\u5f0f\u5b66\u4e60\u52a8\u4f5c\u5f15\u8d77\u7684\u72b6\u6001\u53d8\u5316\u3002\u8bad\u7ec3\u76ee\u6807\u5305\u62ec\u56e0\u679c\u6548\u679c\u9884\u6d4b\u3001\u4efb\u52a1\u805a\u7c7b\u548c\u9636\u6bb5\u8fdb\u5ea6\uff0c\u8ba9Causal Interaction Signal\u771f\u6b63\u643a\u5e26\u7269\u7406\u56e0\u679c\u4fe1\u606f\u3002<\/p>\n

\u4eae\u70b9\u4e8c\uff1a\u53cc\u65f6\u6807\u56e0\u679c\u8bb0\u5fc6\uff0c\u8ba9\u6a21\u578b\u53ef\u4ee5\u8de8\u5c1d\u8bd5\u81ea\u6211\u8fdb\u5316\u3002<\/p>\n

BIT\u4fdd\u8bc1\u5f53\u524d\u6267\u884c\u7684\u8fde\u8d2f\u6027\uff0cPIM\u628a\u4e00\u6b21\u5931\u8d25\u3001\u4e00\u6b21\u4fee\u6b63\u3001\u4e00\u6b21\u6210\u529f\u6c89\u6dc0\u4e3a\u540e\u7eed\u53ef\u68c0\u7d22\u7684\u4e0a\u4e0b\u6587\u3002\u540c\u4e00\u4e2a\u6a21\u578b\u5728repeated attempts\u4e2d\u8d8a\u7528\u8d8a\u5f3a\u3002<\/p>\n

\u4eae\u70b9\u4e09\uff1aIn-Context Policy Injection\uff0c\u8ba9\u51bb\u7ed3\u6a21\u578b\u76f4\u63a5\u6d88\u8d39\u56e0\u679c\u4e0a\u4e0b\u6587\u3002<\/p>\n

\u901a\u8fc7Causal Prompt\u628a\u4efb\u52a1\u3001\u9636\u6bb5\u548c\u4ea4\u4e92\u8bc1\u636e\u6ce8\u5165\u751f\u6210\u6a21\u578b\uff0c\u4e0d\u52a0\u989d\u5916\u566a\u58f0\u3001\u4e0d\u53c2\u4e0eflow loss\uff0c\u53ea\u4f5c\u4e3a\u6761\u4ef6\u5f71\u54cd\u540e\u7eed\u52a8\u4f5c\u3002\u8fd9\u662f\u201c\u4e0d\u8c03\u6743\u91cd\u4e5f\u80fd\u8fdb\u5316\u201d\u7684\u5173\u952e\u8bbe\u8ba1\u3002<\/p>\n

\u548c\u4f20\u7edfTest-Time Training\u76f8\u6bd4\uff0cZeva\u7684\u5dee\u5f02\u5f88\u76f4\u63a5\uff1a<\/p>\n

\u5728RoboCasa365-Atomic5\u4e0a\uff0cZeva\u5e73\u5747\u6210\u529f\u738776.8%\uff0c\u6548\u679c\u6700\u4f18\u3002<\/p>\n

\u66f4\u91cd\u8981\u7684\u662f\u90e8\u7f72\u540e\u7684self-evolution\u66f2\u7ebf\uff1a\u7d2f\u8ba1\u6210\u529f\u7387\u4eceEvolve 1\u768426%\u63d0\u5347\u5230Evolve 4\u768473%\u3002<\/p>\n

\u771f\u5b9e\u5316\u5b66\u5b9e\u9a8c\u5ba4ChemLab-Evo\u4e0a\uff0c\u4e09\u4e2a\u539f\u5b50\u4efb\u52a1\u540c\u6837\u5448\u73b0\u5355\u8c03\u589e\u957f\uff1aPick Up Test Tube\u4ece65%\u5230100%\uff0cPlace Beaker\u4ece25%\u523070%\uff0cPour Water\u4ece30%\u523080%\u3002<\/p>\n

Zeva\u8fd8\u652f\u6301one-shot human demonstration enhancement\uff1a\u4e00\u6b21\u4eba\u7c7b\u6f14\u793a\u521d\u59cb\u5316PIM\u540e\uff0c\u6a21\u578b\u65e0\u9700\u5fae\u8c03\u5373\u53ef\u590d\u73b0\u5173\u952e\u64cd\u4f5c\u3002<\/p>\n

\u91cf\u5316\u7ed3\u679c\u663e\u793a\uff0c\u4eba\u7c7bwarm-up\u5728Place Beaker\u4e0a\u6700\u9ad8\u5e26\u676520\u4e2a\u767e\u5206\u70b9\u7684\u63d0\u5347\uff0c\u5728Pour Water\u4e0a\u5e26\u676515\u4e2a\u767e\u5206\u70b9\u7684\u63d0\u5347\u3002<\/p>\n

\u25b3RoboCasa365\u4e0a\uff0cZeva\u5728\u6a21\u578b\u6743\u91cd\u5b8c\u5168\u51bb\u7ed3\u7684\u60c5\u51b5\u4e0b\uff0c\u7d2f\u8ba1\u6210\u529f\u7387\u4eceEvolve 1\u768426%\u63d0\u5347\u5230Evolve 4\u768473%\u3002\u56fe\u7247\u6765\u6e90\uff1a\u8bba\u6587<\/p>\n

\u25b3\u771f\u5b9e\u5316\u5b66\u5b9e\u9a8c\u5ba4\u4e2d\uff0c\u4e09\u4e2a\u539f\u5b50\u4efb\u52a1\u7d2f\u8ba1\u6210\u529f\u7387\u968f\u5c1d\u8bd5\u91cc\u7a0b\u7891\u63d0\u5347\u3002\u6a21\u578b\u53c2\u6570\u672a\u66f4\u65b0\u3002\u56fe\u7247\u6765\u6e90\uff1a\u8bba\u6587<\/p>\n

\u25b3One-shot\u4eba\u7c7b\u6f14\u793awarm-up\u5bf9\u4e09\u4e2a\u539f\u5b50\u4efb\u52a1\u7684\u5f71\u54cd\u3002\u865a\u7ebf\u4e3a\u6709\u4eba\u7c7b\u6f14\u793a\u521d\u59cb\u5316\uff0c\u5b9e\u7ebf\u4e3a\u7eaf\u81ea\u8fdb\u5316\u3002\u56fe\u7247\u6765\u6e90\uff1a\u8bba\u6587<\/p>\n

Zeva\u7684\u6df1\u5c42\u610f\u4e49\uff0c\u662f\u628a\u5177\u8eab\u6a21\u578b\u7684scaling\u4ece\u201c\u53c2\u6570\u7a7a\u95f4\u201d\u5ef6\u4f38\u5230\u201c\u4e0a\u4e0b\u6587\u7a7a\u95f4\u201d\u3002<\/p>\n

\u4f20\u7edfscaling\u901a\u8fc7\u6269\u5927\u6570\u636e\u3001\u53c2\u6570\u3001\u7b97\u529b\u6765\u63d0\u5347\u6a21\u578b\u80fd\u529b\u3002<\/p>\n

Zeva\u5c55\u793a\u7684\u53e6\u4e00\u79cdscaling\u662f\uff1a\u5728\u90e8\u7f72\u8fc7\u7a0b\u4e2d\u4e0d\u65ad\u589e\u52a0\u53ef\u4f9b\u6a21\u578b\u6d88\u8d39\u7684\u56e0\u679c\u4e0a\u4e0b\u6587\u3002<\/p>\n

\u6bcf\u4e00\u6b21\u5c1d\u8bd5\u90fd\u4f1a\u4ea7\u751f\u65b0\u7684interaction evidence\uff0c\u8fd9\u4e9b\u8bc1\u636e\u88ab\u538b\u7f29\u4e3aCausal Interaction Signal\u5e76\u8fdb\u5165\u53cc\u65f6\u6807\u8bb0\u5fc6\u3002<\/p>\n

\u6a21\u578b\u6bcf\u591a\u4e00\u6b21\u4ea4\u4e92\uff0c\u5c31\u591a\u4e00\u6b21\u5bf9\u5f53\u524d\u7269\u7406\u73af\u5883\u7684\u201c\u9690\u5f0f\u7cfb\u7edf\u8fa8\u8bc6\u201d\u3002<\/p>\n

\u8fd9\u91cc\u7684\u7406\u8bba\u57fa\u7840\u662f\uff1a\u6a21\u578b\u53ef\u4ee5\u628a\u6bcf\u4e2a\u65b0\u573a\u666f\u7684\u73af\u5883\u5c5e\u6027\u4f5c\u4e3a\u9690\u53d8\u91cf\uff0c\u5728\u63a8\u7406\u65f6\u4ece\u8fc7\u53bb\u7684\u52a8\u4f5c-\u6548\u679c\u8bc1\u636e\u4e2d\u63a8\u65ad\u51fa\u6765\u3002<\/p>\n

\u7269\u4f53\u8d28\u91cf\u3001\u5173\u8282\u7ea6\u675f\u3001\u63a5\u89e6\u7279\u6027\u3001\u6469\u64e6\u6761\u4ef6\uff0c\u90fd\u53ef\u4ee5\u901a\u8fc7\u4ea4\u4e92\u4fe1\u53f7\u88ab\u9690\u5f0f\u4f30\u8ba1\u3002\u51bb\u7ed3\u7b56\u7565\u56e0\u6b64\u53ef\u4ee5\u5728\u4e0d\u6539\u53d8\u53c2\u6570\u7684\u60c5\u51b5\u4e0b\uff0c\u4e0d\u65ad\u903c\u8fd1\u5f53\u524d\u7269\u7406\u7cfb\u7edf\u7684\u771f\u5b9e\u52a8\u6001\u3002<\/p>\n

\u66f4\u8fdb\u4e00\u6b65\uff0cZeva\u7684Causal Interaction Signal\u662f\u4e00\u4e2a\u6548\u5e94\u7a7a\u95f4\u4e2d\u7684\u7d27\u51d1\u8868\u793a\uff0c\u76f4\u63a5\u7f16\u7801\u7269\u7406\u6548\u5e94\u4fe1\u606f\u3002<\/p>\n

\u8de8\u4efb\u52a1effect retrieval\u5b9e\u9a8c\u663e\u793a\uff0c\u4e0d\u540c\u7269\u4f53\u3001\u89c6\u89d2\u548c\u4efb\u52a1\u6307\u4ee4\u4e0b\u7684\u7b49\u4ef7\u7269\u7406\u6548\u5e94\uff0c\u4f1a\u5728\u8be5\u7a7a\u95f4\u4e2d\u5f7c\u6b64\u9760\u8fd1\u3002<\/p>\n

\u8fd9\u610f\u5473\u7740\u6a21\u578b\u5b66\u5230\u7684\u662f\u201c\u8fd9\u7c7b\u7269\u7406\u6548\u5e94\u5982\u4f55\u4ea7\u751f\u201d\uff0c\u540c\u65f6\u4e5f\u8986\u76d6\u5177\u4f53\u4efb\u52a1\u7684\u505a\u6cd5\u3002\u8fd9\u79cd\u8868\u793a\u5929\u7136\u652f\u6301\u8de8\u57df\u8fc1\u79fb\uff0c\u4e5f\u8ba9\u201c\u57df\u53d8\u6362\u201d\u6210\u4e3a\u4e00\u4e2a\u53ef\u8ba1\u7b97\u3001\u53ef\u68c0\u7d22\u3001\u53ef\u590d\u7528\u7684\u6280\u672f\u6982\u5ff5\u3002<\/p>\n

\u5982\u679c\u628aVLA\u770b\u4f5c\u201c\u4efb\u52a1\u7406\u89e3\u4e0e\u52a8\u4f5c\u751f\u6210\u201d\u7684\u57fa\u7840\u8bbe\u65bd\uff0c\u4e16\u754c\u6a21\u578b\u770b\u4f5c\u201c\u7ecf\u9a8c\u60f3\u8c61\u201d\u7684\u57fa\u7840\u8bbe\u65bd\uff0c\u90a3\u4e48Zeva\u5b9a\u4e49\u7684ICCL\u5c31\u662f\u201c\u73b0\u573a\u56e0\u679c\u5b66\u4e60\u201d\u7684\u57fa\u7840\u8bbe\u65bd\u3002<\/p>\n

\u5b83\u628a\u90e8\u7f72\u4ece\u201c\u80fd\u529b\u6d88\u8017\u201d\u53d8\u6210\u201c\u80fd\u529b\u79ef\u7d2f\u201d\uff0c\u628a\u6bcf\u4e00\u6b21\u5931\u8d25\u4ece\u566a\u58f0\u53d8\u6210\u53ef\u68c0\u7d22\u7684\u56e0\u679c\u8bc1\u636e\u3002<\/p>\n

\u5982\u679cICCL\u6210\u7acb\uff0c\u5177\u8eab\u667a\u80fd\u7684scaling\u5c06\u591a\u51fa\u4e00\u6761\u72ec\u7acb\u4e8e\u53c2\u6570\u89c4\u6a21\u7684\u66f2\u7ebf\uff1a\u4ea4\u4e92\u6b21\u6570\u8d8a\u591a\uff0c\u56e0\u679c\u4e0a\u4e0b\u6587\u8d8a\u4e30\u5bcc\uff0c\u6a21\u578b\u5bf9\u5f53\u524d\u7269\u7406\u57df\u7684\u8fa8\u8bc6\u8d8a\u51c6\uff0c\u4efb\u52a1\u6210\u529f\u7387\u8d8a\u9ad8\u3002\u8fd9\u624d\u662f\u201c\u57df\u53d8\u6362\u201d\u4f5c\u4e3a\u65b0\u8303\u5f0f\u7684\u771f\u6b63\u542b\u4e49\u3002<\/p>\n

\u6e05\u534eAIR\u4e0e\u57df\u53d8\u6362\u8054\u624b\u53d1\u5e03\u7684Zeva\uff0c\u91cd\u65b0\u5b9a\u4e49\u4e86\u5177\u8eab\u6a21\u578b\u7684\u81ea\u8fdb\u5316\uff1a<\/p>\n

\u4e0d\u8c03\u6743\u91cd\uff0c\u4e5f\u80fd\u8d8a\u7528\u8d8a\u5f3a\uff1b\u4e0d\u91cd\u65b0\u8bad\u7ec3\uff0c\u4e5f\u80fd\u4ece\u4e00\u6b21\u4eba\u7c7b\u6f14\u793a\u4e2d\u5b66\u4f1a\u65b0\u64cd\u4f5c\u3002<\/p>\n

\u57df\u53d8\u6362\u5f00\u542f\u7684\uff0c\u53ef\u80fd\u662f\u4e00\u6761\u65b0\u7684scaling\u8def\u5f84\uff1a\u8ba9\u673a\u5668\u4eba\u5728\u771f\u5b9e\u7269\u7406\u4e16\u754c\u4e2d\uff0c\u81ea\u5df1\u5b66\u4f1a\u56e0\u679c\uff0c\u81ea\u5df1\u5b8c\u6210\u8fdb\u5316\u3002<\/p>\n

\u53c2\u8003\u94fe\u63a5\uff1a[1]\u8bba\u6587\uff1aZeva: In-Context Causal Interaction Memory for Embodied Action Generalization[2]\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/air-embodied-brain.github.io\/Zeva\/<\/p>\n

\u5173\u4e8e\u6e05\u534eAIR\u4e0e\u57df\u53d8\u6362<\/p>\n

\u6e05\u534eAIR\uff08Institute for AI Industry Research, Tsinghua University\uff09\u662f\u6e05\u534e\u5927\u5b66\u9762\u5411\u7b2c\u56db\u6b21\u5de5\u4e1a\u9769\u547d\u7684\u56fd\u9645\u5316\u3001\u667a\u80fd\u5316\u3001\u4ea7\u4e1a\u5316\u7684\u7814\u7a76\u673a\u6784\uff0c\u88ab\u89c6\u4e3a\u4e2d\u56fdAI\u4ea7\u5b66\u7814\u7ed3\u5408\u7684\u7b2c\u4e00\u68af\u961f\u3002\u57df\u53d8\u6362\u662f\u6e05\u534eAIR\u9762\u5411\u7269\u7406\u667a\u80fd\u81ea\u8fdb\u5316\u65f6\u4ee3\u5b75\u5316\u7684\u516c\u53f8\uff0c\u7531\u5218\u4e91\u65b0\u6559\u6388\u548c\u66f9\u5a77\u6559\u6388\u53d1\u8d77\uff0c\u672c\u6b21\u53d1\u5e03\u7684Zeva\u662f\u6e05\u534eAIR\u5728\u5177\u8eab\u667a\u80fd\u65b9\u5411\u4ece\u524d\u6cbf\u7814\u7a76\u8d70\u5411\u4ea7\u4e1a\u843d\u5730\u7684\u91cd\u8981\u6807\u5fd7\u3002<\/p>\n

\u5218\u4e91\u65b0<\/p>\n

\u6e05\u534e\u5927\u5b66\u667a\u80fd\u4ea7\u4e1a\u7814\u7a76\u9662\u526f\u9662\u957f\u3001\u535a\u5bfc\u3001\u4e07\u56fd\u6570\u636e\u51a0\u540d\u6559\u6388\u3001\u9996\u5e2d\u7814\u7a76\u5458\uff0c\u56fd\u5bb6\u7ea7\u9ad8\u5c42\u6b21\u4eba\u624d\u3001IEEE Fellow<\/p>\n

\u524d\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u9996\u5e2d\u7814\u7a76\u4e3b\u7ba1\uff0c\u957f\u671f\u4ece\u4e8bAIoT\u3001\u7aef\u4fa7\u667a\u80fd\u7814\u7a76<\/p>\n

\u83b7\u5f97\u4e86MobiCom\u3001MobiSys\u3001SenSys\u7b49\u591a\u4e2a\u56fd\u9645\u9876\u7ea7\u5b66\u672f\u4f1a\u8bae\u5956\u52b1\uff0c\u4ee5\u53caCCF\u79d1\u6280\u8fdb\u6b65\u4e00\u7b49\u5956\u7b49\u5956\u9879\u3002<\/p>\n

\u66f9\u5a77<\/p>\n

\u6e05\u534e\u5927\u5b66\u6559\u6388\uff0c\u5317\u4eac\u5e02\u9ad8\u5c42\u6b21\u5f15\u8fdb\u4eba\u624d<\/p>\n

\u524d\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u9996\u5e2d\u7814\u7a76\u4e3b\u7ba1\uff0c\u7814\u7a76\u65b9\u5411\u4e3a\u8fb9\u7f18\u667a\u80fd\u3001\u5177\u8eab\u667a\u80fd\u7b49<\/p>\n

\u83b7\u9876\u7ea7\u4f1a\u8bae\u6700\u4f73\u8bba\u6587\u5956\u56db\u6b21\uff0c\u5de5\u4f5c\u5165\u9009ACM\/Microsoft\u7814\u7a76\u4eae\u70b9\u7b49\uff0c\u5176\u4e2dACM\u7814\u7a76\u4eae\u70b9\u7531\u56fe\u7075\u5956\u83b7\u5f97\u8005David Patterson\u4eb2\u81ea\u64b0\u6587\u63a8\u8350\u3002<\/p>\n

\n

\u7248\u6743\u58f0\u660e\uff1a<\/strong><\/p>\n

\u672c\u535a\u5ba2\u90e8\u5206\u5185\u5bb9\u4e3a\u8f6c\u8f7d\u6587\u7ae0\uff0c\u65e8\u5728\u5206\u4eab\u6709\u4ef7\u503c\u7684\u4fe1\u606f\uff0c\u7248\u6743\u5f52\u539f\u4f5c\u8005\u6240\u6709\u3002<\/p>\n

\u8f6c\u8f7d\u4ec5\u4e3a\u4e2a\u4eba\u5b66\u4e60\u4e0e\u4ea4\u6d41\u76ee\u7684\uff0c\u4e0d\u5bf9\u6587\u7ae0\u89c2\u70b9\u8d1f\u8d23\uff0c\u4ea6\u4e0d\u7528\u4e8e\u4efb\u4f55\u5546\u4e1a\u7528\u9014\u3002<\/p>\n

\u5982\u6d89\u53ca\u7248\u6743\u95ee\u9898\uff0c\u8bf7\u8054\u7cfb\u672c\u4eba\u5220\u9664\u3002<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"

Zeva\u56e2\u961f \u6295\u7a3f<\/p>\n

\u6a21\u578b\u4e0d\u66f4\u65b0\uff0c\u80fd\u529b\u5374\u6301\u7eed\u589e\u957f\u3002<\/p>\n

\u8fd9\u662fZeva\u6700\u53cd\u76f4\u89c9\u7684\u5730\u65b9\uff0c\u4e5f\u662f\u6e05\u534eAIR\u4e0e\u57df\u53d8\u6362\u8054\u624b\u7ed9\u51fa\u7684\u65b0\u7b54\u6848\u3002<\/p>\n

\u5728\u591a\u4e2a\u5178\u578b\u5177\u8eab\u57fa\u51c6\u4efb\u52a1\u4e0a\uff0c\u5c06\u51bb\u7ed3\u6a21\u578b\u7684\u7d2f\u8ba1\u6210\u529f\u7387\u4ece26%\u63d0\u5347\u523073%\uff1b\u5728\u771f\u5b9e\u5316\u5b66\u5b9e\u9a8c\u5ba4\uff0c\u673a\u68b0\u81c2\u5728\u591a\u6b21\u5c1d\u8bd5\u4e2d\u8d8a\u7528\u8d8a\u7a33\u3002<\/p>\n

\u66f4\u5173\u952e\u7684\u662f\uff0c\u4e00\u6b21\u4eba\u7c7b\u6f14\u793a\u5c31\u80fd\u8ba9\u6a21\u578b\u201c\u5b66\u4f1a\u201d\u4e00\u4e2a\u65b0\u64cd\u4f5c\u3002<\/p>\n

Zeva\u662f\u9996\u4e2a\u5b9e\u73b0In-Context Causal Learning\uff08ICCL\uff09\u7684\u5177\u8eabWAM\u3002<\/p>\n","protected":false},"author":1,"featured_media":7504,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[8,1165,904],"class_list":["post-7508","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-xwzx","tag-8","tag-1165","tag-air"],"_links":{"self":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/7508","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/comments?post=7508"}],"version-history":[{"count":0,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/7508\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media\/7504"}],"wp:attachment":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media?parent=7508"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/categories?post=7508"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/tags?post=7508"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}